Visualizing Learning Systems
While machine learning systems such as artificial neural networks have been employed in a wide range of application domains in the past, their dynamics have often evaded traditional methods of analysis. Throughout this project, the author took an NKS-inspired approach of experimentation and graphical exploration towards the analysis of learning methods, with particular emphasis on the visual properties of their basins of attraction. First, a traditional example of the phenomenon of basins of attraction, Newton’s root finding method, was considered. The aim was to seek complex and interesting behavior through directed experiments. Next, attractors in Hopfield neural networks were investigated through visualization. Finally, the same methodology was applied to search for elementary classification behavior in simple NKS-style sequential substitution systems.
(April 20, 2004)